What Is an AI Visibility Audit? A Practical Checklist
By VouchUpdated 6 min read
An AI visibility audit is a one-off, structured check of how your brand appears in AI answers and, more usefully, why. Tracking tells you that ChatGPT stopped naming you at the trust question. An audit tells you whether the cause is a blocked crawler, a page that buries its answer, missing reviews on the sites assistants cite, or prices that disagree from one site to the next.
AI search visibility defines the metrics, and how to track brand mentions in AI search covers ongoing measurement. This is the diagnostic you run when you first start, or when the numbers move and you need to know why.
Step 1: Collect the questions buyers ask
Pull 20 to 30 real questions from sales calls, support tickets and your keyword data, and cover the whole conversation, not just the opener: Discovery, Shortlist, Objection, Trust, Value and Decision. The buyer conversation prompts give you a set for each stage.
Step 2: Run them and log the answers
Ask each question in ChatGPT, Perplexity, Gemini and Google (for AI Overviews and AI Mode), keeping the follow-ups in the same chat because answers change with context. For each answer, log whether you were named, cited or recommended, which rivals appeared at which question, and which sources the assistant relied on. The column list is in the tracking guide. Keep the source list: steps 4 and 5 use it.
Step 3: Check the technical doors
An assistant can only cite what it can reach.
- robots.txt. OpenAI's crawler documentation says sites that opt out of OAI-SearchBot are not shown in ChatGPT search answers, while GPTBot, the training crawler, is a separate setting. Perplexity says PerplexityBot surfaces sites in its search results. For Google's AI features, a page must be indexed and eligible for a snippet, according to Google's AI features guide.
- Firewalls and CDNs. OpenAI's help centre says a site's host or CDN must allow traffic from OpenAI's published search-bot IP addresses for the site to be eligible for ChatGPT search. Aggressive bot protection can block an assistant your robots.txt allows.
- Indexing. ChatGPT search sometimes uses other search providers, and OpenAI's help page links Microsoft's privacy statement for them, so check Bing Webmaster Tools alongside Search Console.
- Rendering. Text that only appears after scripts run, or sits behind a login, is harder for crawlers to read and quote.
The free AI crawler checker covers most of this in one go: robots.txt for 14 AI and search crawlers, llms.txt, server-rendered text and noindex settings, with a plain fix for each problem.
Step 4: Audit the pages that should be cited
For each question in step 1, find the page on your site that should answer it, then check:
- Does it state the answer clearly near the top, under a heading that matches the question?
- Does it give specifics: numbers, dates, names, comparisons? Statistics and cited sources are among the changes the GEO research found lift visibility in AI answers.
- Is it dated and current?
- Where step 2 shows a rival's page cited instead, what does that page answer that yours does not?
If no page answers the question, that is the finding. For a page that does, the free AI readiness page audit checks the parts a program can: whether AI crawlers may read it, whether it answers up front, its headings, its date and its structured data.
Step 5: Audit your third-party coverage
Assistants build answers from comparison articles, reviews, forums and directories, not only from your domain. Go through the sources from step 2. Are you listed? Are you described accurately, with current prices and products? Are your reviews recent? How Perplexity chooses sources shows how much one assistant relies on comparison pages. Missing or stale profiles on the sites assistants already cite are often the cheapest gap to close.
Step 6: Find the stage where you drop out
The most useful output is not a score but the stage where you disappear, because each stage points to a different fix:
| Where you drop out | What it usually points to |
|---|---|
| Discovery | Missing from the category's comparison and "best of" coverage |
| Shortlist | No clear comparison page, or weak differentiation |
| Objection | Known complaints left unanswered on your own pages |
| Trust | Thin reviews, warranty or after-sales proof |
| Value | Unclear or inconsistent pricing |
| Decision | No obvious reason or place to buy |
Killer questions explains how to find the question that ends your run and what to publish to survive it.
Step 7: Set the baseline and the cadence
Write down where you stand, pick the fixes, and audit again on a schedule. Monthly is a sensible default; sooner if you ship major content or an assistant in your category changes. Without a baseline you cannot tell whether any of the work moved anything.
Common questions
How long does a manual audit take? Plan on about a day for a first pass of 20 to 30 questions across four assistants, plus the crawler, page and third-party checks. Careful logging matters more than the tooling.
Can I audit without paid tools? Yes. The steps above are the audit, and the crawler checker is free. Tools add consistent sampling, history and scale; they do not change what you are looking for.
What is the difference between an audit and tracking? Tracking measures the answers over time. An audit is a one-off look at the causes behind them: access, pages, coverage and the stage where you drop out.
Let Vouch run steps 2 and 6 for you
Vouch asks your buyers' questions every day across seven assistants and runs six-question Threads on ChatGPT, Claude, Gemini and Perplexity. For each tracked question it shows the stage where your brand was dropped, who replaced it, the sources the assistant cited and what to publish next. Our methodology explains how each assistant is asked and scored.
Vouch is invite-only for now, and every invitation starts with a 14-day free trial: join the waitlist.
